Core Concepts
Incorporating infrastructure data significantly improves the accuracy and noise robustness of collaborative perception for autonomous driving, particularly when infrastructure acts as the primary agent (infra-centric approach) in specific scenarios like intersections.
Stats
Incorporating infrastructure data improves 3D detection accuracy by up to 10.87%.
Infra-centric CP increases accuracy by up to 42.53% compared with vehicle-centric CP.
In V2XSet-I, the detection range of z is converted from z ∈[−3, 1] to z ∈[−5, −1].
In V2X-Sim, the detection range of z is converted from z ∈[−3, 2] to z ∈[−8.5, −3.5].